I am Head of Machine Learning R&D at a big multinational corporation. Well, I’m not. But let’s pretend I were. If I were, my “corporate context” may be one where there is one central R&D unit (which is where I live), and several business units. Let’s assume these business units are Robotics, Energy Generation and Distribution, Transport and Logistics, and Manufacturing. In addition, given that everything these days is computerized, networked, software-ized, Cyber Security is the elephant in the room. That is, it does not live in its own business unit, but it lurks in the background of all other BUs.
Now, as said Head of ML R&D, I may do some machine learning research without any application in mind. But the bulk of my time would (and should, I’d say) probably be spent in the context of some application. For example, I may consider developing new ML methods for robotics (such as learning how to grasp things); for energy distribution (such as how weather influences energy load across my grid, given that some power generation methods are weather dependent); or for cyber security (such as how to distinguish good from bad activity in my networks). Almost certainly, I will not have the resources to simply do everything I or my team members would like to do. I will have to prioritize.
For prioritizing, I could go by the principle of subsidiarity. That is, I might only take on those topics that can or should not be addressed by a business unit directly. This will probably include (1) topics that do not (yet) have an immediate business case, and (2) topics that are relevant across business units. So I need to find a way of grouping topics into the following buckets:
Bucket 1: No immediately assignable business case (yet).
Bucket 2: Relevant across business units.
Bucket 3: Matches one of my business units, and business case exists.
(1) and (2) I should do in my central R&D unit; (3) I should try to delegate to business units.
Finding topics to delegate to business units
I will delegate topics that match a business unit, and that have a business case. In order to identify such topics, I use Mergeflow’s Topic Triage tool. This tool lets you plot topics against each other, and compare their size and growth across science and business. The Topic Triage tool uses real-time data from across the web and other sources, analyzes and displays them, and provides access to the data (in the real solution, not in our screenshots here below).
My plan is to identify topics that show momentum across science as well as business, and then delegate those topics to individual BUs.
First, I plotted “machine learning for robotics” against “machine learning for autonomous transport and logistics”, and so on. The result is the following chart:
This chart shows how my topics unfold over time, across science, tech blogs, venture investments, and industry. Green are high-growth topics (= more than 10% CAGR); yellow is moderate growth (1-10% CAGR), and red is negative growth (CAGR < 0%). So, for example, “ML for Manufacturing” is high-growth but not very big yet, whereas “ML for Robotics” is negative growth but very big.
Now we need to find topics to delegate, i.e. those that show momentum across science and business. Here they are (the legend shows that I left out patents; I did this because patents lag behind in time with their publication, and this makes comparisons with the other types of information a bit tricky):
All three show momentum across science and business. Time to delegate.
Finding topics for my R&D unit, and what to do with them
Well, now all that’s left are “ML for Power Generation and Distribution”…
…and “ML for Cybersecurity”:
As you can see, “ML for Power Generation and Distribution” is very science-heavy. Not much business happening there (yet). This means that it lands in my Bucket No. 1 (cf. above).
“ML for Cybersecurity” does show momentum across science and business, but we said in the beginning that it is an “elephant in the room” topic that applies across all our business units. This puts it into Bucket No. 2 (again, cf. above for bucket definitions).
Notice that I would probably treat the two topics differently, given their different signature in the Topic Triage tool. For the science-heavy “ML for Power Generation and Distribution”, I could start with building prototypes, based on the research that’s published, and based on what I can come up with in my own unit. Ideally, I could then talk to my Power Generation BU about possible business models. But importantly, doing research and prototypes will be me.
For “ML for Cybersecurity” I would most likely proceed differently, and start from the business side. Of course, my corporation is forward-thinking and does not suffer from NIH (= Not Invented Here) Syndrome. This means that I will have no issues if I talk to one of the venture-funded companies in the area about possible collaborations. This will include my technology experts talking to their technology experts about what makes their technologies special. Based on the outcome of this, I will then make a roadmap for connecting outside technologies with my business units.